Optimal design for the estimation of variance components (Q3774769)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 4034916
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Optimal design for the estimation of variance components |
scientific article; zbMATH DE number 4034916 |
Statements
Optimal design for the estimation of variance components (English)
0 references
1988
0 references
estimation of variance components
0 references
multifactor random effects model
0 references
balanced design
0 references
minimax designs
0 references
Hölder's inequality
0 references
cross classification
0 references
method of unweighted squares of means
0 references
numerical methods
0 references
0.803341805934906
0 references
0.7998649477958679
0 references
0.7930540442466736
0 references
0.7930540442466736
0 references
A general model II of an m-factorial cross classification with m factors \(F_ 1,...,F_ m\) is considered. From the j-th factor \(s_ j\geq 2\) \((j=1,...,m)\) levels are included in the experiment. A typical cell is then \(i=(i_ 1,...,i_ m)\) with \(n_ i>0\) observations each. At least one \(n_ i\) has more than one observation. All main and interaction effects up to order m-1 are included in the model and as well as the error terms assumed to be normally and independently distributed with zero means and unknwon variances. Under these assumptions it is shown that for fixed values \(s_ j\) \((j=1,...,m)\) and when the variance components are estimated by the method of unweighted squares of means, the variance of each estimator will be minimized for equal cell numbers \(n_ i=const\). The optimal selection can therefore be considered for all \(n_ i\) equal. But two problems arise. First, the designs are locally optimum, i.e. for any variance component the optimum choice of the \(s_ j\) that minimizes the variance of the estimator of this component depends on the unknown parameters. Secondly, for any fixed set of values of the unknown parameters, no single choice of the \(s_ j\) minimizes the variance of all estimators. Analytical solutions of the problem are not obtained. Numerical methods based on a complete enumeration are mentioned as successful.
0 references